In testing for the number of components in a mixture, it is well known that regularity conditions do not hold for minus twice the log likelihood ratio to have its usual asymptotic null distribution of chi-squared. In an attempt to overcome some of the problems, it has been proposed in the literature that a prior distribution be placed on the mixing proportions. We show here that the usual regularity conditions still do not hold.
|Number of pages||4|
|Journal||Journal of the Royal Statistical Society. Series B: Statistical Methodology|
|Publication status||Published - 1987|
- mixture model
- number of components
- prior distribution
- likelihood ratio test